So you decided to adopt an algorithm. Good for you, you go to the RL algorithm shelter and you have three to choose from. Daisy is a Q-Learning mixed breed with some Neural network of an unknown percentage. After looking at some pics, she looks a little more Q-Learning, than a Neural net, but those breeds usually go together pretty well to make a powerful algorithm
Next you have Rocky who is a pure bred policy Gradient Network. You’ve looked at the paperwork and the documentation and he is IN FACT a pure-bred. No mixing in any Q nets, actor-critics, or any hand jammed values in Rocky, he is a Policy Gradient and nothing else!
Last you’ve got Lucy who is a 75% Actor-Critic method and a 25% Deep Q-Network. The code base mentions that you can use Lucy alone but she may get awfully lonely working without her best friend Champ. They come as a pair and it’s not recommended to separate the two, but there’s no API to combine both algorithms together. You’ve got to adopt both, but Champ’s just gonna hang out with Lucy. They were never siblings by birth but they grew up so close that you’ll deal with lots of issues if you separate the two. Who do you pick?!?